Research

Paper

TESTING March 02, 2026

Uniform-in-time concentration in two-layer neural networks via transportation inequalities

Authors

Arnaud Guillin, Boris Nectoux, Paul Stos

Abstract

We quantify, uniformly over time and with high probability, the discrepancy between the predictions of a two-layer neural network trained by stochastic gradient descent (SGD) and their mean-field limit, for quadratic loss and ridge regularization. As a key ingredient, we establish T p transportation inequalities (p $\in$ {1, 2}) for the law of the SGD parameters, with explicit constants independent of the iteration index. We then prove uniform-in-time concentration of the empirical parameter measure around its mean-field limit in the Wasserstein distance W 1 , and we translate these bounds into prediction-error estimates against a fixed test function $Φ$. We also derive analogous concentration bounds in the sliced-Wasserstein distance SW 1 , leading to dimension-free rates.

Metadata

arXiv ID: 2603.01842
Provider: ARXIV
Primary Category: cs.NE
Published: 2026-03-02
Fetched: 2026-03-03 04:34

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.01842v1</id>\n    <title>Uniform-in-time concentration in two-layer neural networks via transportation inequalities</title>\n    <updated>2026-03-02T13:19:59Z</updated>\n    <link href='https://arxiv.org/abs/2603.01842v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.01842v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>We quantify, uniformly over time and with high probability, the discrepancy between the predictions of a two-layer neural network trained by stochastic gradient descent (SGD) and their mean-field limit, for quadratic loss and ridge regularization. As a key ingredient, we establish T p transportation inequalities (p $\\in$ {1, 2}) for the law of the SGD parameters, with explicit constants independent of the iteration index. We then prove uniform-in-time concentration of the empirical parameter measure around its mean-field limit in the Wasserstein distance W 1 , and we translate these bounds into prediction-error estimates against a fixed test function $Φ$. We also derive analogous concentration bounds in the sliced-Wasserstein distance SW 1 , leading to dimension-free rates.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.NE'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='math.PR'/>\n    <published>2026-03-02T13:19:59Z</published>\n    <arxiv:primary_category term='cs.NE'/>\n    <author>\n      <name>Arnaud Guillin</name>\n      <arxiv:affiliation>LMBP</arxiv:affiliation>\n    </author>\n    <author>\n      <name>Boris Nectoux</name>\n      <arxiv:affiliation>LMBP</arxiv:affiliation>\n    </author>\n    <author>\n      <name>Paul Stos</name>\n      <arxiv:affiliation>LMBP</arxiv:affiliation>\n    </author>\n  </entry>"
}